News Seagate Ships 20TB HAMR HDDs Commercially, Increases Shipments of Mach.2 Drives

[Admin]

Seagate brings future closer with commercial HAMR hard drives, expansion of dual-actuator HDD evaluation program.

Seagate Ships 20TB HAMR HDDs Commercially, Increases Shipments of Mach.2 Drives : Read more

 

[Sledgehamma]

Is there any word when the end consumer can buy those 20TB drives?

 

[Kamen Rider Blade]

When can we get Dual Actuator Drives as a consumer?

 

[TheJoker2020]

"Is there any word when the end consumer can buy those 20TB drives? "

"When can we get Dual Actuator Drives as a consumer? "

I can tell you now that you simply do not want one of these as a consumer, at least not in their current form.!

From the Data Sheet:

"14 TB of capacity available as two independently addressable, 7 TB logical units"

Essentially this is seen by the OS (assume W10 for the average consumer) as two separate drives.!

This is fine if you are running (for example) a RAID5 setup as you can use 4 of these drives, seen as 8 drives and you can double the performance without doubling the physical quantity of drives, there will of course be caveats, especially for reliability as the failure of one drive will often be two "logical" drives that fail at the same moment.!

There are a number of better solutions available right now for the "consumer" market. If and when this technology arrives and is seen and operates as a single drive to the end user rather than the end user essentially running the two "logical" drives as RAID0 in a single drive, that would be good and I look forward to it, if it arrives soon enough to be useful.

Addendum, here is the link to the data sheet.

https://www.seagate.com/files/www-content/datasheets/pdfs/exos-2x14-DS2015-3-2007GB-en_GB.pdf

 

[spongiemaster]

Seagate's first-to-market dual actuator technology

Nope. You would think Seagate knew this was false since they bought the company that beat them to this title by over 25 years. Conner released the Chinook all the way back in 1994.

 

[TheJoker2020]

Nope. You would think Seagate knew this was false since they bought the company that beat them to this title by over 25 years. Conner released the Chinook all the way back in 1994.

Seagate would simply claim that it OWNS Conner and it's Patents, many of which were used by Seagate well after Conner was just a memory.

https://en.wikipedia.org/wiki/Conner_Peripherals

 

[kyzarvs]

Nope. You would think Seagate knew this was false since they bought the company that beat them to this title by over 25 years. Conner released the Chinook all the way back in 1994.

The single-spindle bit as quoted a couple of times in the article is the 'new' bit...

 

[kyzarvs]

"Is there any word when the end consumer can buy those 20TB drives? "

"When can we get Dual Actuator Drives as a consumer? "

I can tell you now that you simply do not want one of these as a consumer, at least not in their current form.!

From the Data Sheet:

"14 TB of capacity available as two independently addressable, 7 TB logical units"

Essentially this is seen by the OS (assume W10 for the average consumer) as two separate drives.!

This is fine if you are running (for example) a RAID5 setup as you can use 4 of these drives, seen as 8 drives and you can double the performance without doubling the physical quantity of drives, there will of course be caveats, especially for reliability as the failure of one drive will often be two "logical" drives that fail at the same moment.!

There are a number of better solutions available right now for the "consumer" market. If and when this technology arrives and is seen and operates as a single drive to the end user rather than the end user essentially running the two "logical" drives as RAID0 in a single drive, that would be good and I look forward to it, if it arrives soon enough to be useful.

Addendum, here is the link to the data sheet.

https://www.seagate.com/files/www-content/datasheets/pdfs/exos-2x14-DS2015-3-2007GB-en_GB.pdf

So you JBOD or RAID 1 them to get a 14TB logical drive? JBOD wouldn't see any preformance benefit, RAID1 has more performance, but higher chance of failure.

 

[spongiemaster]

So you JBOD or RAID 1 them to get a 14TB logical drive? JBOD wouldn't see any preformance benefit, RAID1 has more performance, but higher chance of failure.

If you used software RAID, JBOD would give you 14TB, RAID 1, which would mirror the logical drives, would only give you 7TB. RAID 0 would give you 14TB. The drive has the ability to stream from both actuators at the same time to maximize throughput if needed, so there is no need to use RAID 0.

As mentioned above, this drive is not designed for home use. This is intended for servers with massive capacity drive arrays where the drive capacity is filled and getting hammered constantly. In this use case, with 2 independent head assemblies you basically have 2 drives inside the same enclosure which could theoretically double the I/O's and throughput vs having just one actuator in the enclosure. This is an unrealistic scenario for home use especially at this capacity. You'll get far better performance buying a smaller NVME drive for your boot drive/apps and a 2nd 2TB NVME drive for your games or whatever.

 

[Mr5oh]

You'll get far better performance buying a smaller NVME drive for your boot drive/apps and a 2nd 2TB NVME drive for your games or whatever.

There's those of us out there that 2TB isn't enough for game / data drive. I'm using WD Gold drives in my Gaming machine for this reason. Just saying you might see these large capacity drives in home computers more often than you think.

 

[Endymio]

I can tell you now that you simply do not want one of these as a consumer,....if you are running (for example) a RAID5 setup as you can use 4 of these drives, seen as 8 drives and you can double the performance without doubling the physical quantity of drives there will of course be caveats, especially for reliability as the failure of one drive will often be two "logical" drives that fail at the same moment.!

A) You're confusing Raid 5 with Raid 0/10 configurations.
B) Seagate forcing the drive to appear as two logical units, then the consumer (using Raid 0) forces it back to a single logical drive = zero total impact on reliablity. It's still just a single drive, and a single point of failure.

Now, potentially, you could have a failure of a single actuator, which would allow the other to continue functioning, a benefit you would lose with the Raid 0 setup. However, this benefit wouldn't exist if Seagate presented the unit as a single logical drive, so again there's no reliability downside to this approach.

 

[Endymio]

The drive has the ability to stream from both actuators at the same time to maximize throughput if needed, so there is no need to use RAID 0.

Only true for high-queue depth data warehousing-type applications. For a consumer-level drive, Raid 0 will still be faster. You can't use both actuators at once unless a particular file is split between both (logical) drives.

 

[TheJoker2020]

A) You're confusing Raid 5 with Raid 0/10 configurations.

No. I originally stated RAID 5/6/50/60 whereupon you could use multiple drives (each drive being 2 logical drives) and you could use them in these kinds of RAID setups for the added performance. So as to not confuse people too much I removed the extra RAID options that this drive would be suitable for, perhaps I should have left them in.

B) Seagate forcing the drive to appear as two logical units, then the consumer (using Raid 0) forces it back to a single logical drive = zero total impact on reliability. It's still just a single drive, and a single point of failure.

No. The failure rate would increase because there are many potential points of failure within a drive, and this having two actuators doubles the chances of failure at this point, likewise the controller will be new and the failure rate here is unknown currently.

Until there is largescale usage of these drives and we know "how" they were used, their overall failure rate is simply unknown, until then we can merely speculate.

As per the average user using one of these drives with both logical drives in a software RAID-0 setup also adds the potential for the end user to screw up by say, formatting one of the drives. Consumer oriented products that will I hope arrive soon would ideally have the drive operate as RAID-0 internally so that the end user cannot screw it up, better still, have a jumper on the drive to optionally change it between a single logical drive or two logical drives. What though would the end user plug the drive into.? Will SAS become a thing on desktop boards.?

Of course this kind of technology could be extended to have more than two actuators in the future, and there is no reason why this could not also be incorporated into a dual-actuator SSHD.

 

[TheJoker2020]

Only true for high-queue depth data warehousing-type applications. For a consumer-level drive, Raid 0 will still be faster. You can't use both actuators at once unless a particular file is split between both (logical) drives.

Which it would be if you setup the two logical drives as RAID-0.

 

[Endymio]

No. I originally stated RAID 5/6/50/60 whereupon you could use multiple drives (each drive being 2 logical drives) and you could use them in these kinds of RAID setups for the added performance. So as to not confuse people too much I removed the extra RAID options..

No. The mistake you made in your original posting (and perpetuated here) is the belief that Raid 5 doubles performance (post #1) while halving the number of logical drives (this post). Raid 5 requires a minimum of THREE drives, reducing the number of logical drives by 1/3, (or 1/4, or 1/5, etc.) but never, ever by half. And in typical Raid 5 (and 5+, 50, etc) access patterns, it does boost performance (sometimes by much more than 200% in a wide array. ) However, in a typical consumer-level access pattern, such configurations may actually reduce performance.

Raid 0 and similar configurations are what you are thinking of, which (very roughly speaking) double performance while halving the logical drive count.

No. The failure rate would increase because there are many potential points of failure within a drive, and this having two actuators doubles the chances of failure at this point. Until there is largescale usage of these drives and we know "how" they were used, their overall failure rate is simply unknown, until then we can merely speculate.

Again, your logic is faulty, on two separate points. Firstly, we don't have to "speculate" on a drive's reliability: we use the MTTF/MTBF data the manufacturer provides through testing. But much more importantly, you were claiming that two actuators in a single drive was less reliable than having those same two actuators in two separate drives, when the exact opposite is true. Since this apparently isn't as obvious as it should be, I'll elaborate with a (very simplified) example:

Seagate dual-actuator drive: one controller board, one spindle drive motor, two actuators = 4 failure points.
Two single-actuator drives: two controller boards, two spindle drive motors, two actuators = 6 failure points.

Assuming equally reliable subcomponents, the dual-actuator solution will have a higher MTTF (though obviously the failure may impact both logical drives, rather than simply one).

As per the average user using one of these drives with both logical drives in a software RAID-0 setup also adds the potential for the end user to screw up by say, formatting one of the drives.

Err, what? A user can format one single 20TB drive just as easily as they can format one of the individual drives in a 2x10TB array -- even more easily, in fact, as most RAID firmware and software setups add additional confirmation checks.

 

[Endymio]

Which it would be if you setup the two logical drives as RAID-0.

Did you read the post before you replied to it? The OP stated Raid-0 wasn't necessary, as it would be equally fast to use it as two native logical drives. I pointed out that Raid 0 is necessary for general consumer access patterns.

 

[spongiemaster]

There's those of us out there that 2TB isn't enough for game / data drive. I'm using WD Gold drives in my Gaming machine for this reason. Just saying you might see these large capacity drives in home computers more often than you think.

Then buy another SSD or learn how to uninstall games. With SSD's dropping below $100 per terabyte, no one would recommend a RAID 0 mechanical drive array over SSD's for pretty much any speed sensitive home use case. For regular large scale file storage where speed is largely irrelevant, a mechanical drive is the way to go, though certainly not in a RAID 0 array.

 

[spongiemaster]

Did you read the post before you replied to it? The OP stated Raid-0 wasn't necessary, as it would be equally fast to use it as two native logical drives. I pointed out that Raid 0 is necessary for general consumer access patterns.

In one of the promotional videos for the drive someone mentioned optimization for streaming applications that made it sound like there was a mode where the 2 LUNs could be combined to make one 500+MBps stream. I can't find any reference to this in the documentation, so it looks like the max throughput can only be achieved through 2 independent 250MBps data streams. In a FAQ, Seagate says it would be possible to make the drive look like one LUN and have the drive load balance on its own, but that there would have to be sufficient market demand for them to develop such a firmware.

 

[spongiemaster]

Again, your logic is faulty, on two separate points. Firstly, we don't have to "speculate" on a drive's reliability: we use the MTTF/MTBF data the manufacturer provides through testing.

MTBF is meaningless when applied to a single drive. A drive with an MTBF of 1 million hours is obviously not expected to last 114 years of continual usage on average. MTBF only starts to become statically relevant when your drive count gets into the 1000's, and the rating is really intended to apply to the entire manufacturing run of the model.

 

[Endymio]

MTBF is meaningless when applied to a single drive. A drive with an MTBF of 1 million hours is obviously not expected to last 114 years of continual usage on average. MTBF only starts to become statically relevant when your drive count gets into the 1000's

So many misconceptions here. The two largest are:

a) The notion that sampling data doesn't apply to individual cases is an utter misunderstanding of basic statistics, akin to saying that the difference in auto-accident vs. lightning-strike fatality rates don't apply to you, because you're only one single person, who has yet to be killed by either.
b) MTBF calculations (or MTTF, or AFR, depending on the Mgfr) are premised upon on a bathtub-curve assumption valid only under that time period. No, a MTBF of 1M hours doesn't imply 114 years of constant use, rather it allows you to calculate for an individual drive the expected failure probability only during the trough of that curve -- the so-called "random failure" period.

 

[wirefire]

"Is there any word when the end consumer can buy those 20TB drives? "

"When can we get Dual Actuator Drives as a consumer? "

I can tell you now that you simply do not want one of these as a consumer, at least not in their current form.!

From the Data Sheet:

"14 TB of capacity available as two independently addressable, 7 TB logical units"

Essentially this is seen by the OS (assume W10 for the average consumer) as two separate drives.!

This is fine if you are running (for example) a RAID5 setup as you can use 4 of these drives, seen as 8 drives and you can double the performance without doubling the physical quantity of drives, there will of course be caveats, especially for reliability as the failure of one drive will often be two "logical" drives that fail at the same moment.!

There are a number of better solutions available right now for the "consumer" market. If and when this technology arrives and is seen and operates as a single drive to the end user rather than the end user essentially running the two "logical" drives as RAID0 in a single drive, that would be good and I look forward to it, if it arrives soon enough to be useful.

Addendum, here is the link to the data sheet.

https://www.seagate.com/files/www-content/datasheets/pdfs/exos-2x14-DS2015-3-2007GB-en_GB.pdf

Yeah this is a wonderful idea. put 2 separately addressable drives in one unit. then use raid 5 on the devices. if you just HAPPEN to lose a physical device and not one of the addressable units. you lose the array.

I wouldn't trust these in a datacenter further than I could throw them... if drives are going to use this kind of technology and they dont present to the controlled like their single actuator cousins. you are looking at potentially needing an entirely new breed of raid controllers to manage these kinds of devices. (yeah yeah... raid 6 would in theory work but with these drives the raid 6 calculations and overhead give you raid 5 redundancy.)

 

[spongiemaster]

So many misconceptions here. The two largest are:

a) The notion that sampling data doesn't apply to individual cases is an utter misunderstanding of basic statistics, akin to saying that the difference in auto-accident vs. lightning-strike fatality rates don't apply to you, because you're only one single person, who has yet to be killed by either.
b) MTBF calculations (or MTTF, or AFR, depending on the Mgfr) are premised upon on a bathtub-curve assumption valid only under that time period. No, a MTBF of 1M hours doesn't imply 114 years of constant use, rather it allows you to calculate for an individual drive the expected failure probability only during the trough of that curve -- the so-called "random failure" period.

a)It's the most basic of statistical concepts that the smaller the data sample size you have, the less likely it is to accurately represent the entire pool. I'm not interested debating such a basic concept.
b)Industry standard for hard drive manufactures is that MTBF applies to the service life of a hard drive which is typically 5 years assuming the drive did not exceed the rated daily powered on hours (24 hours for server drives). So, theoretically if you replaced your drive every 5 years, you could statistically expect a failure every 114 years. But again, because you would only be working with 23 drives over that course of time, whatever failure rate you ended up with would be worthless as an accurate representation of the entire field beyond a coincidence.

 

[Endymio]

a)It's the most basic of statistical concepts that the smaller the data sample size you have, the less likely it is to accurately represent the entire pool.

Again, so many misconceptions. The "sample size" in this case is the number of drives (actually, the number of drive-hours) the manufacturer uses to calculate the MTBF. The larger that sample size, the more accurately MTBF(sample) represents MTBF(all drives). No matter how many drives you possess, the chance of any single drive failing is based upon that MTBF. You're confused over the meaning of sampling.

You stated that statistical values are "meaningless" (your word) when applied to individuals. I won't go into the mathematics of why this is utterly fallacious (if you're interested, google "sigma" for a start), but to place things in perspective, consider two brain surgeons, both of whom have performed thousands of operations. One has seen 90% of his patients die on the table, the other less than 1%. By your reasoning, you shouldn't care which surgeon you choose. After all, you're just one person, getting one surgery, and thus those failure rates are an absolute meaningless representation of your own chance of surviving.

So -- you going to choose Surgeon B?

I'm not interested debating such a basic concept.

It's not open to debate.. not since the year 1812 or so.

 

[Endymio]

(yeah yeah... raid 6 would in theory work but with these drives the raid 6 calculations and overhead give you raid 5 redundancy.)

Finally, someone who understands at least the basics of Raid. Your points are valid, however there are a few factors you're not considering.

a) Such a configuration would still be near Raid-6 redundancy, because most failures are single-sector reads anyway, and some "whole drive" failures would still allow you to read the alternate LUN.
b) If you're a large datacenter with many wide arrays, then if you want "true" Raid 6 redundancy (or true Raid-5, without the R6 overhead), you can structure the LUNs into separate arrays
c) For JBOD warehousing, none of this matters.

Remember, the rationale for the drive is that you get 20TB for "close to" the form factor and price of a 10TB drive. Assuming that, there are plenty of applications where this will outweigh the negatives of the dual-LUN presentation.

 

[spongiemaster]

No matter how many drives you possess, the chance of any single drive failing is based upon that MTBF. You're confused over the meaning of sampling.

You stated that statistical values are "meaningless" (your word) when applied to individuals. I won't go into the mathematics of why this is utterly fallacious (if you're interested, google "sigma" for a start), but to place things in perspective, consider two brain surgeons, both of whom have performed thousands of operations.

I admire how confident you are in your incorrectness while feeling the need to continually try to insult me and accuse me of not knowing what I am talking about. The following is pulled from a Western Digital data sheet for one of their enterprise drives:

Projected values. Final MTBF and AFR specifications will be based on a sample population and are estimated by statistical measurements and acceleration algorithms under typical operating conditions, workload 220TB/year and temperature 40C. Derating of MTBF and AFR will occur above these parameters, up to 550TB writes per year and 60°C ambient (65°C device temp). MTBF and AFR ratings do not predict an individual drive’s reliability and do not constitute a warranty

https://documents.westerndigital.co...c500-series/data-sheet-ultrastar-dc-hc550.pdf


So Western Digital has put in writing that MTBF does not mean anything for an individual drive, but please, continue acting like I'm the idiot for saying that earlier.